Analysis of Dogs’ Sleep Patterns Using Convolutional Neural Networks

This paper describes a system for the automatic analysis of sleeping patterns in kenneled dogs, developed as an indicator of their welfare. Addressing the gap in adequate automatic analysis systems for dogs, especially when using low-quality video, the system uniquely combines convolutional neural networks (CNNs) with classical data processing methods. It processes video footage from simple web or security cameras, even with very low quality, and is capable of detecting multiple dogs in a frame. The core tasks of the system involve localizing dogs within each frame and classifying their state as either awake or asleep. An end-to-end architecture, specifically Faster R-CNN ResNet101, is used for detection and classification, followed by a post-processing module that corrects potential errors in localization and classification. This automated solution provides an efficient and accurate way to quantify sleep parameters—such as total sleep amount, sleep interval count, and sleep interval length—for large volumes of video data, which would otherwise be a tedious and error-prone manual task. The research highlights the potential of neural networks to revolutionize animal behavior and welfare science.